Diffractive deep neural networks with hardware-software co-design

Abstract

A multipath deep diffractive neural network comprises a first optical path for performing a first task and a second optical path for performing a second task. The second task is different than the first task. The multipath deep diffractive neural network further comprises an overlap optical path where the first optical path and the second optical path overlap. The multipath deep diffractive neural network comprises one or more optical elements that are configured to create a multipath optical neural network that performs a plurality of different tasks using multi-task machine learning.

What is claimed is: 
     
         1 . A multipath deep diffractive neural network comprising:
 a first optical path for performing a first task;   a second optical path for performing a second task, wherein the second task is different than the first task;   an overlap optical path where the first optical path and the second optical path overlap;   one or more optical elements; and   wherein the one or more optical elements are configured to create a multipath optical neural network that performs a plurality of different tasks using multi-task machine learning.   
     
     
         2 . The multipath deep diffractive neural network of  claim 1 , wherein the one or more optical elements are configured to select a choice of orientation from a group consisting of 0, 45, and 90 degrees. 
     
     
         3 . The multipath deep diffractive neural network of  claim 1 , wherein the one or more optical elements comprise nanomaterials. 
     
     
         4 . The multipath deep diffractive neural network of  claim 1 , wherein the one or more optical elements comprise one or more of passive beam splitters and active spatial light modulators. 
     
     
         5 . The multipath deep diffractive neural network of  claim 1 , wherein the multipath deep diffractive neural network is configured to perform reflection harvesting. 
     
     
         6 . The multipath deep diffractive neural network of  claim 5 , further comprising:
 one or more shared diffractive layers located on the overlap optical path, wherein the one or more shared diffractive layers are configured to be used within the first task and the second task.   
     
     
         7 . The multipath deep diffractive neural network of  claim 6 , further comprising:
 an optical element positioned at an end of the overlap optical path, wherein the optical element is configured to transmit a signal along the first optical path and reflect the signal along the second optical path.   
     
     
         8 . The multipath deep diffractive neural network of  claim 7 , further comprising:
 one or more first diffractive layers positioned along the first optical path, wherein the one or more first diffractive layers are configured to be used within the first task.   
     
     
         9 . The multipath deep diffractive neural network of  claim 8 , further comprising:
 one or more second diffractive layers positioned along the second optical path, wherein the one or more second diffractive layers are configured to be used within the second task.   
     
     
         10 . The multipath deep diffractive neural network of  claim 1 , further comprising:
 a third optical path for performing a third task, wherein the third task is different than the first task and the second task.   
     
     
         11 . A method for using a multipath deep diffractive neural network comprising:
 causing a signal to be provided to an overlap optical path where a first optical path and a second optical path overlap, wherein:
 the first optical path is configured to perform a first task, 
 the second optical path is configured to perform a second task, 
   wherein the second task is different than the first task, and
 one or more optical elements are configured to create a multipath optical neural network that performs a plurality of different tasks using multi-task machine learning. 
   
     
     
         12 . The method as recited in  claim 11 , wherein the one or more optical elements are configured to select a choice of orientation from a group consisting of 0, 45, and 90 degrees. 
     
     
         13 . The method as recited in  claim 11 , wherein the one or more optical elements comprise nanomaterials. 
     
     
         14 . The method as recited in  claim 11 , wherein the one or more optical elements comprise one or more of passive beam splitters and active spatial light modulators. 
     
     
         15 . The method as recited in  claim 11 , wherein the multipath deep diffractive neural network is configured to perform reflection harvesting. 
     
     
         16 . The method as recited in  claim 11 , wherein one or more shared diffractive layers are located on the overlap optical path, wherein the one or more shared diffractive layers are configured to be used within the first task and the second task. 
     
     
         17 . The method as recited in  claim 16 , wherein an optical element positioned at an end of the overlap optical path transmits a signal along the first optical path and reflect the signal along the second optical path. 
     
     
         18 . The method as recited in  claim 17 , wherein one or more first diffractive layers are positioned along the first optical path, wherein the one or more first diffractive layers are configured to be used within the first task. 
     
     
         19 . The method as recited in  claim 18 , wherein one or more second diffractive layers are positioned along the second optical path, wherein the one or more second diffractive layers are configured to be used within the second task. 
     
     
         20 . The method as recited in  claim 11 , further comprising:
 a third optical path for performing a third task, wherein the third task is different than the first task and the second task.